Turing as a Functional Programming Language

Cervantes · **Posted:** Sun May 21, 2006 9:53 pm

(Partly.)

What is a functional programming language?

Wikipedia wrote:

Functional programming is a programming paradigm that treats computation as the evaluation of mathematical functions. Functional programming emphasizes the definition of functions rather than the implementation of state machines, in contrast to procedural programming, which emphasizes the execution of sequential commands. A purely functional program does not use mutation: rather than modifying state to produce values (as is done in imperative programming), it constructs new values from (but does not overwrite) existing values.

There is no uniform agreement on what constitutes functional programming or a functional programming language. Often considered important are first-class functions (including anonymous functions), closures, and recursion. Other common features of functional programming languages are continuations, Hindley-Milner type inference systems, non-strict evaluation (i.e. "laziness") or explicit suspensions, and monads.

another function that takes one parameter

first-class values

Functions can be anonymous. We don't need to give every function we make a name.
Functions can be passed to other functions as values. This is really what this tutorial is about.

Functions as parameters

code:

function findRoot (low, high, incriment : real) : real
% Find the first root of the function f(x) = x^2 - 4
% starting from 'low' and ending at 'high'.
var x := low
loop
exit when x > high
% The rounding here is to avoid things like
% 0.000000000023345 not equalling 0
if round ((x ** 2 - 4) / incriment) = 0 then
result x
end if
x += incriment
end loop
% No root was found for the function f(x) = x^2 - 4
% between x = low and x = high
% Return a signal that no root was found.
result minint
end findRoot

put "The first root that I found between -10 and 10 for the function f(x) = x**2 - 4 is: ", findRoot (-10, 10, 0.1)

But we can do better. Our 'findRoot' function isn't very useful if it only deals with one hardcoded function, f(x) = x^2 - 4. To make it really useful, our 'findRoot' function should take another function as a parameter. This parameter will be the function whose root will be found.
Here we go. I'm really going to do it. Passing functions as parameters. Just let me put on my wizard robe and pick up my magic wand.
I'll bet some of you have been reading this and thinking none of it will actually be possible in Turing. Never say "never".

code:

function findRoot (f : function f (x : real) : real, low, high, incriment : real) : real
% Find the first root of the function f(x)
% starting from 'low' and ending at 'high'.
var x := low
loop
exit when x > high
% The rounding here is to avoid things like
% 0.000000000023345 not equalling 0
if round (f (x) / incriment) = 0 then
result x
end if
x += incriment
end loop
% No root was found for the function f(x) = x^2 - 4
% between x = low and x = high
% Return a signal that no root was found.
result minint
end findRoot

function f (x : real) : real
result x ** 2 - 4
end f
put "The first root that I found between -10 and 10 for the function f(x) = x**2 - 4 is: ", findRoot (f, -10, 10, 0.1)

Press F1. ... Success!

There you have it. Functions in Turing can be passed as parameters.

Now some notes:

The function parameter in our 'findRoot' function is 'f'. That is, our function is called 'f'. This does not mean that the parameter type -- function f (x : real) : real -- has to use 'f' as the function name. The following is also valid, for example:

code:

function findRoot (f : function g (x : real) : real, low, high, incriment : real) : real
There is an alternative syntax to do this. It's essentially the same, but without the initial 'f :' in 'f : function ...'. For example,

code:

function findRoot (function f (x : real) : real, low, high, incriment : real) : real
To the best of my knowledge (which, with regard to functionalism in Turing, has been put together in one night), we cannot anonymously create functions. That is, we must define our function f(x) as per usual before calling our 'findRoot' function, then pass f(x) into 'findRoot'.

Another (more useful) example to digest

here

code:

module Calculus
export slope, area
function slope (f : function f (x : real) : real, x : real) : real
const accuracy := 0.00000001
result (f (x + accuracy) - f (x - accuracy)) / (accuracy * 2)
end slope
function area (f : function f (x : real) : real, low, high, width : real) : real
if high - low <= width then
result (high - low) * f ((high + low) / 2)
else
result width * f (low + width / 2) + area (f, low + width, high, width)
end if
end area
end Calculus

You can test this module with

code:

function f (x : real) : real
result x ** 2
end f

put "The slope of 'f' at x = 5 is ", Calculus.slope (f, 5)
put "The area under 'f' from x = 0 to x = 5 is ", Calculus.area (f, 0, 5, 0.01)

Or an even more fun test, which just might show you a pattern in all this calculus:

code:

function f (x : real) : real
result x ** 2
end f
put " x | slope of f(x) at x"
put "-------+--------------------"
for x : -10 .. 10
put " ", x : 3, " | ", Calculus.slope (f, x)
end for

output:

x | slope of f(x)
-------+--------------------
-10 | -20
-9 | -18
-8 | -16
-7 | -14
-6 | -12
-5 | -10
-4 | -8
-3 | -6
-2 | -4
-1 | -2
0 | 0
1 | 2
2 | 4
3 | 6
4 | 8
5 | 10
6 | 12
7 | 14
8 | 16
9 | 18
10 | 20

slope of f(x) = 2x, if f(x) = x^2.

If you wish to further your knowledge of Functional Programming, you'll need to learn a new language. Haskell is a purely functional programming language. O'Caml is a great language (I personally recommend it) that fully supports functional programming, but does not limit you to that programming paradigm.

I hope you've enjoyed this lesson in functional programming, in mathematical functions, and in Calculus. Feel free to ask questions.

Delos · **Posted:** Mon May 22, 2006 12:15 pm

Great tut. This sort of idea (passing constructs as parameters) is a very, very powerful one that people too often overlook. It might be interesting right now to refresh people's memories that one is also capable of creating arrays of functions/procedures; which when coupled with this tut, could produce some intriguing results.

Now a little questions:

Cervantes wrote:

3 To the best of my knowledge (which, with regard to functionalism in Turing, has been put together in one night), we cannot anonymously create functions. That is, we must define our function f(x) as per usual before calling our 'findRoot' function, then pass f(x) into 'findRoot'.

Care to define this concept of anonmity a little better? Things were well explained until you hit that comment and then everything went up in a well-woven handbasket.

As for the Calculus - I'm guessing your disturbingly indepth knowledge of calc is a result of some 'advanced' courses you took (AP?), since I know that I didn't learn even the basics of integration in high school (though some others I know did).

As a parting request, how about adding some 'problems to attempt' at the end - get people to apply these ideas in practical settings. I know a lot of the more advanced coders around here would appreciate this largely.

Cervantes · **Posted:** Mon May 22, 2006 1:14 pm

Delos wrote:

Cervantes wrote:

3 To the best of my knowledge (which, with regard to functionalism in Turing, has been put together in one night), we cannot anonymously create functions. That is, we must define our function f(x) as per usual before calling our 'findRoot' function, then pass f(x) into 'findRoot'.

Care to define this concept of anonmity a little better? Things were well explained until you hit that comment and then everything went up in a well-woven handbasket.

Certainly.

Anonymous Functions
(using my own made up Turing syntax, because this doesn't really work in Turing)

var my_string := "hello world"

code:

put length("hello world")

is the same as

code:

var my_string := "hello world"
put length (my_string)

This much should be obvious.

Now, if a function is truly a first-class value, it too shouldn't need a name (be bound to a handle, or assigned to a variable) to exist. Instead of defining f(x) and then passing our function f into our function 'Calculus.slope', we should be able to anonymously create our function to give to 'Calculus.slope', just as we arbitrarily give the string literal "hello world" to the 'length' function.

Here's an example, with my own home-cooked Turing syntax (note: this code will NOT execute!)

code:

put "The slope of f(x) = x ** 3 + 2 when x = 5 is ", Calculus.slope (lambda {[x : real] x ** 3 + 2}, 5)

The syntax I've made up here uses "lambda" (from lambda calculus) to anonymously define a function, where the parameters to that function are contained with brackets [], and the whole anonymous function is contained within curly braces {}.

So here I've defined the function that we are giving to 'Calculus.slope' right in the middle of calling 'Calculus.slope'. No need to give it a name; we're never going to use such a boring curve as y = x^3 + 2 later in the program. It's just like put length ("hello world")

Anonymous Functions in Other Languages

O'Caml

Ocaml:

fun x -> x ** 3 + 2

'x' is the parameter, and everything after the "->" is the body of the function. This anonymous function could be directly inserted into another function that accepts a function as a parameter, such as a Riemann Sum function.

Ruby also has functional aspects. Anonymous functions in Ruby come in the form of blocks, or Proc objects..

Ruby:

module Calculus
def self.slope(f, x)
accuracy = 0.0001
(f.call(x + accuracy) - f.call(x - accuracy)) / (accuracy * 2)
end
end
puts Calculus.slope(lambda {|x| x ** 2}, 5)

Questions?

_________________

Delos wrote:

As for the Calculus - I'm guessing your disturbingly indepth knowledge of calc is a result of some 'advanced' courses you took (AP?), since I know that I didn't learn even the basics of integration in high school (though some others I know did).

That's right. AP for the win!

Delos wrote:

As a parting request, how about adding some 'problems to attempt' at the end - get people to apply these ideas in practical settings. I know a lot of the more advanced coders around here would appreciate this largely.

I'll see what I can do. There's some more examples coming (examples that aren't purely for math!), so maybe I'll divert some of them to being "exercises".

wtd · **Posted:** Mon May 22, 2006 2:11 pm

This is fantastic stuff. Keep it coming, pirate boy. Smile

Cervantes · **Posted:** Tue May 23, 2006 6:42 pm

Here's some more examples, some of which deviate from the realm of math and calculus that the first examples were limited to.

TURuby

IntArray

code:

class IntArray
export var value_at
var value_at : array 1 .. 1000 of int
var n := 0
end IntArray

var first_arr, second_arr : ^IntArray
new IntArray, first_arr

You'll notice that I used a big honking static array, rather than a flexible array, because Turing cannot export a dynamic array. You'll also notice that I called the internal array 'value_at'. Seems a strange name for an array, doesn't it? But look -- it allows us to do something like this (once we've initialized the object):

code:

first_arr -> value_at (3)

This gets the third element from the array, and it looks just like a method call. Pretty sweet, huh? Now, onwards and upwards, my friends! We're going to add a whole bunch of really cool methods to this IntArray class.

First, I will define a helper method, push. This accepts one argument and pushes it onto the top of the array. It is defined like this:

code:

proc push (value : int)
n += 1
value_at (n) := value
end push

You'll only see the push method appear in the code later, when I define the map and delete_if methods.

The first interesting method is create, which is used to initialize the array. It takes two parameters. The first is the number of elements the array will store. The second parameter, however, is a function f that takes one integer argument and returns another integer. The idea is that the nth element of the array will be set to f(n). Here is the code:

code:

class IntArray
export var value_at, create
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial) % This reads quite nicely: "create 4 factorials"

What is in our array, now that we've created it with a factorial function? If you've done the math, you'd know that the array holds [1, 2, 6, 24]. If we want to print that, we need to iterate over every element in the array. That's where our next method, each, comes in.

each is our iteration method. It takes each element of our array and performs an action on it. The action that it performs is determined by the procedure that you pass to each. Here is the code:

code:

class IntArray
export var value_at, create, each
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
proc each (do : procedure do (x : int))
for i : 1 .. n
do (value_at (i))
end for
end each
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial)

proc print (x : int)
put x, " " ..
end print
first_arr -> each (print)

output wrote:

1 2 6 24

What if we want to iterate through each element of our array, but still have the index by our side? We'll make a new method, each_with_index, whose procedure takes two parameters, x and i, which represent the element and the index, respectively. Here is the code:

code:

class IntArray
export var value_at, create, each_with_index
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
proc each_with_index (do : procedure do (x, i : int))
for i : 1 .. n
do (value_at (i), i)
end for
end each_with_index
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial)

proc print_table (x, i : int)
put i, " | ", x
end print_table
first_arr -> each_with_index (print_table)

The next method is very special. It is the inject method. It is so special that wtd wrote an entire tutorial on it, entitled The Power of Inject. (It's a Ruby tutorial, but inject is a common method in many languages.)

Inject takes two parameters: an integer to start with and a function f. The function takes two parameters, a and b. Basically, inject starts with an initial value and takes the first element of the array, passing both of these into the function f. It stores the result of that function in its local variable. It then moves on to the second element of the array, passing it's local variable (which is f(starting_value, first_element)) and the second element to the function. This process is repeated until we reach the end of the array. The code might better illustrate this:

code:

class IntArray
export var value_at, create, inject
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
function inject (start : int, f : function f (a, b : int) : int) : int
var ans := start
for i : 1 .. n
ans := f (ans, value_at (i))
end for
result ans
end inject
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial)

function sum (a, b : int) : int
result a + b
end sum
function maximum (a, b : int) : int
if a > b then
result a
else
result b
end if
end maximum
put "The sum of the entries in first_arr is ", first_arr -> inject (0, sum)
put "The largest number in first_arr is ", first_arr -> inject (0, maximum)

Inject is a powerful method indeed, but it's power is limited by the functions we give it.

The next method I've made is the map method. map takes each element of our array and applies a function to it, returning a new array. It "maps" the function onto the array.

code:

class IntArray
export var value_at, create, map
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
proc push (value : int)
n += 1
value_at (n) := value
end push
function map (f : function f (x : int) : int) : ^IntArray
var new_arr : ^IntArray
new IntArray, new_arr
for i : 1 .. n
new_arr -> push (f (value_at (i)))
end for
result new_arr
end map
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial)

function perfect_square (x : int) : int
result x * x
end perfect_square
put "The squares of the first four factorials are: " ..
first_arr -> map (perfect_square) -> each (print)
put ""

The final method I've got is delete_if. This method takes one parameter: a function. This function takes one parameter: an integer. This function returns a boolean. The delete_if method will return a new array that is a duplicate of the current array, except all the elements of our current array who satisfy the function will be deleted. Here is the code:

code:

class IntArray
export var value_at, create, delete_if
var value_at : array 1 .. 1000 of int
var n := 0
proc create (num_of_elements : int, f : function f (x : int) : int)
n := num_of_elements
for i : 1 .. num_of_elements
value_at (i) := f (i)
end for
end create
proc push (value : int)
n += 1
value_at (n) := value
end push
function delete_if (f : function f (x : int) : boolean) : ^IntArray
var new_arr : ^IntArray
new IntArray, new_arr
for i : 1 .. n
if f (value_at (i)) = false then
new_arr -> push (value_at (i))
end if
end for
result new_arr
end delete_if
end IntArray

var first_arr : ^IntArray
new IntArray, first_arr

function factorial (n : int) : int
if n < 2 then
result 1
else
result n * factorial (n - 1)
end if
end factorial
first_arr -> create (4, factorial)

proc print (x : int)
put x, " " ..
end print
function even (x : int) : boolean
result x mod 2 = 0
end even
put "The only elements of first_arr that even are: " ..
first_arr -> delete_if (even) -> each (print)
put ""

And that's all the methods I've reimplimented into Turing.

Conclusion

anonymous functions

code:

first_arr -> delete_if (lambda {[x : int] x mod 2 = 0}) -> each (lambda {[x : int] put x, " " ..})

to the non-anonymous approach,

code:

proc print (x : int)
put x, " " ..
end print
function even (x : int) : boolean
result x mod 2 = 0
end even
first_arr -> delete_if (even) -> each (print)

More Information

give Ruby a try (in your browser! No installing anything)

inject

Ruby:

[1, 2, 4, 10, 67].inject(0) {|a, b| a + b}

You can download the entire Turing IntArray class and the demonstration of it in the attached source file.

Feel free to ask questions.

Delos · **Posted:** Tue May 23, 2006 8:47 pm

/blown away

That was pretty damn sweet. These concepts are utterly brilliant, and I can see why OOP is so bloody powerful. Many a heartfelt thanks for taking the time to implement such code in Turing so that us uninitiates can learn a thing or two.

To that effect...you have an error! Check out your delete_if method. It pushes on false as opposed to true as your description says it should...

Anonymous fcns eh...interesting concept...

Cervantes · **Posted:** Tue May 23, 2006 9:12 pm

Delos wrote:

To that effect...you have an error! Check out your delete_if method. It pushes on false as opposed to true as your description says it should...

There is no error. Wink

It pushes on false. Only if the function returns false is that element pushed into the new array. Conversely, if the function returns true, that element is not pushed into the new array, and it is thus 'deleted' (though it still lives on in the current array).

If I had instead used a method like delete_at (index : int), then I would have called that whenever the given function f returned true.

Anyone have any suggestions for the next part? I could give the array of procedure example for a GUI widget, but that's kind of boring.

md · **Posted:** Tue May 23, 2006 9:58 pm

Cervantes... you scare me... turing as a functional language (sorta)? How do you even think of these things, let alone actually try?

'Course you can do all of these tricks in other languages too (C, C++, Pascal); though still without the anonymous functions.

wtd · **Posted:** Wed May 24, 2006 10:15 am

Cornflake wrote:

Cervantes... you scare me... turing as a functional language (sorta)? How do you even think of these things, let alone actually try?

'Course you can do all of these tricks in other languages too (C, C++, Pascal); though still without the anonymous functions.

You can't do anonymous functions in C++? Wink

http://www.boost.org/doc/html/lambda.html

Cervantes · **Posted:** Wed May 24, 2006 11:23 am

Cornflake wrote:

Cervantes... you scare me... turing as a functional language (sorta)? How do you even think of these things, let alone actually try?

I was actually looking through the Turing Help Manual to help TheOneTrueGod out when I saw this very thing done. Look under 'paramDeclarations', or something like that (Well, not you, cornflake).

md · **Posted:** Wed May 24, 2006 12:27 pm

I see... and wtd thanks for pointing out that I'm wrong... now I get to play with more things (hopefully) Very Happy

One place where function pointers are rather useful in graphics libraries is when you have a list of sprites. Each sprite might have all the same data in a record to control what and where to draw it, but they might all have different functions controlling how they move; or different collision functions. Instead of having to rely on the person using hte sprite library to call all the correct procedures you can just have two function pointers in your sprite record. One that's called every frame to move and update the sprite, and one that's called on collisions.

There was a sprite library for Pascal on Mac (actually _the_ library) that worked this way and it worked really well.

wtd · **Posted:** Wed May 24, 2006 4:54 pm

Once you realize the expressive power of fucntional programming, you begin to realize how short-sighted Turing is for using keywords for things like IO.

Imagine a world where "put" is a function.

code:

some_int_array -> each (put)

But instead, we have to wrap put in a function.

GlobeTrotter · **Posted:** Wed May 24, 2006 7:30 pm

If "put" was a function, what would it return? I understand get being a function. Do you mean "put" as a procedure?

Delos · **Posted:** Wed May 24, 2006 7:40 pm

No, a function! Think about it...imagine being able to do this:

code:

foo_arr -> each (put ( root_with_base (3)));

On the fly output - this would be useful in debugging sits. I'm sure wtd had more intricate thoughts in mind though, I'm just delving here Wink

.

wtd · **Posted:** Thu May 25, 2006 12:50 pm

Delos wrote:

No, a function! Think about it...imagine being able to do this:

code:

foo_arr -> each (put ( root_with_base (3)));

Without even realizing it, you have described partial function application. It is one of the more powerful concepts in fucntional programming. Smile

	Computer Science Canada Programming C, C++, Java, PHP, Ruby, Turing, VB Username: Password: Register
Wiki Blog Search Turing Chat Room Members